Free resource · no sign-up

The AI Discovery Playbook

How to walk into a logistics company, learn how it really runs, and turn what you find into recommendations people will act on.

Book an AI discovery call

0. Our six rules
1. Before you walk in
2. Questions by role
3. Flows to observe
4. How to observe
5. Readiness check
6. Ranking what you find
7. Recommendation one-pager
8. Timeline and deliverables
Appendix A. Interview note sheet
Appendix B. Flow-step capture sheet

What this is: the field guide we use inside a company during the first 2 to 4 weeks of a fractional Chief AI Officer engagement: how we find out what to do before recommending anything. Use it as is, adapt it, or use it to check whoever you hire.

Listen first. Watch the work. Count it. Then recommend, smallest useful step first.

0. Our six rules

Rule	What it means in practice
1. Listen before you propose	The first two weeks are for learning the business, not pitching tools. Sit with the people doing the work. Do not bring a solution to the first meeting. You earn the right to recommend by showing you understood.
2. Answer "what's in it for me?" for each person	Every recommendation must fix a pain someone named. Write down, for each stakeholder, what they personally get (less re-keying, fewer late calls, a cleaner report). If nobody's day gets better, the business will not care, however clever the tool is.
3. Make governance visible	Do not just say AI will be used responsibly. Show it: a one-page AI use policy, a list of every AI tool and who owns it, named human approvals for money, safety and pricing, and a simple log people can see.
4. Pilot fast, measure business results	Small pilot, real work, 2 to 6 weeks. Measure what the business already tracks (quote turnaround, hours per load, invoice cycle time, missed follow-ups), not model accuracy. Scale what works; stop what doesn't, and say so.
5. Bring the skeptics in	The experienced person who says "we've always done it this way" usually knows where the real exceptions are. Put them on the pilot, let them shape the rules, and credit them in public. Respect experience; show how AI supports it. Pair newer staff with veterans so learning goes both ways.
6. Translate for the CEO	Speak in money, time, risk and customers, never in models. Don't oversell: AI is useful, not magic, and one broken promise costs more trust than three quiet wins earn. Celebrate small wins and name the people who made them work.

1. Before you walk in (checklist)

☐ Role charter agreed in writing: what the AI lead owns, decides, and does not do. Who you report to.
☐ Contract and NDA cover every entity you will look at (sister companies are often separate legal entities).
☐ Intake questionnaire answered (systems, goals, budget, decisions that must stay human).
☐ The CEO has introduced you to the team and said why you're there, in their own words. Do not start interviews before this.
☐ A client-owned folder exists for your notes (their Drive or SharePoint, not yours).
☐ Read everything public and everything the client has given you. Don't ask questions their own plan already answers.
☐ List the stakeholders: name, role, what they're measured on, what they probably fear, what they'd gain.
☐ Book the interviews: 30 to 45 minutes each, at their desk where possible.

Stakeholder map (fill in)

Name / role	Measured on	Biggest pain (their words)	Likely worry about AI	What's in it for them	Supporter / neutral / skeptic

2. Questions by role

Ask about today, not about AI. Use their terms. Ask for an example ("show me the last one") rather than the general case. Use the interview note sheet (Appendix A). End every interview with the magic-wand question.

The magic-wand question (ask everyone, last): "If you could wave a magic wand and make one part of your week disappear or get easier, what would it be?" Then: "What would you do with that time?" Patterns across people are your shortlist.

CEO / owner

What are the top three business goals for the next 12 months, and how will you know you've hit them?
Which numbers do you look at every week? Where do they come from, and do you trust them?
Where do you think money or time leaks today? What keeps you up at night?
What has the company already tried with technology or AI? What worked, what didn't, and why?
Which decisions must always stay with a person (prices, credit, carrier approval, payments, safety, hiring)?
Who decides on tools and vendors, and up to what amount? Who else needs to agree?
Who on the team will be most excited about this, and who most worried?
What would make this engagement an obvious success at 90 days?

Sales / reps / customer service

Walk me through the last lead you handled, from first contact to booked (or lost).
Where do leads come from (web form, phone, email, load board, referrals)? Which ones do you trust?
How long does a quote take you? What slows it down: missing details, looking up rates, waiting on someone?
How do you know who to follow up with today? What falls through the cracks?
Which systems do you have open all day? Where do you type the same thing twice?
How do customers want to hear about their load, and how much time goes on "where's my truck?"

Operations / dispatch / carrier sales

Show me how a load goes from booked to covered to delivered. Who touches it at each step?
How do you find and vet carriers (or assign drivers)? How long does it take on a normal load, and on a hard one?
How do status updates and exceptions reach you: phone, text, ELD message, email? Is any of it logged?
What happens when something breaks (late, breakdown, no-show, wrong paperwork)? Who decides what next?
How much of your day is check calls and chasing paperwork?
What do you track on a whiteboard, spreadsheet or in your head because the system doesn't do it?

Billing / back office / finance

How do PODs and delivery documents come back, and how long until an invoice goes out?
Do you factor or bill direct? What gets an invoice rejected or delayed?
How are carrier or driver payments checked before they go out? Who approves?
Where is revenue missed (accessorials, detention, storage, rebills)?
Which reports do you build by hand each week or month, and how long do they take?

Safety / compliance

Where do driver qualification files, insurance certificates and expiry dates live?
How do you know, before a load goes out, that the driver, truck and carrier are cleared?
How are incidents, claims and corrective actions recorded and followed up?
Who can stop work, and how is that recorded? (Never automate this decision away.)
Do you record calls? Which states do you call into? Any consent or notice rules you follow?

IT / systems / vendors

List every system: TMS, ELD/telematics, CRM, accounting, phone/call tracking, email suite (Google or Microsoft 365), document tools, load boards.
Who administers the PCs, browsers and email accounts? Who holds the admin logins?
Which systems have an API or export? Is it read-only or can it write? Is there a user activity or audit log?
Which outside vendors (software, AI, marketing, call tracking) are under contract? Is there a written scope, milestones and acceptance criteria? Who owns the accounts, code and data?
Is multi-factor sign-in on for email, TMS and banking? Has there been a security incident or fraud event?
Is anyone pasting customer or driver data into public AI tools today?

3. Flows to observe

Map the money path end to end, plus the two support flows. At each step, record the eight things in the capture sheet (Appendix B). Most automation value sits in the handoffs, re-keying and waits between steps, not inside one system.

Step	What to watch for	Typical signals of waste
1. Lead	Where leads land, how fast someone responds, how they're logged and assigned	Leads in personal inboxes; no first-response time; duplicates across tools
2. Quote	Details collected, rate lookup, approval, how the quote is sent	Back-and-forth for missing dims/weights; rates from memory; quote not logged
3. Book	Tender or rate confirmation received, load created in the TMS	Retyping the email or PDF into the TMS
4. Dispatch / cover	Carrier or driver found, vetted, assigned, confirmed	Many calls per load; vetting done by hand; capacity emails unread
5. Track	Status updates, check calls, customer updates, exceptions	Manual check calls; "where's my truck?" calls; exceptions not logged
6. Deliver	POD, photos, OS&D, delivery confirmation	PODs chased by text; photos lost; damage found late
7. Invoice	Documents matched, invoice built and sent (or sent to the factor)	Days between delivery and invoice; missing accessorials
8. Collect / pay	Payments received, carrier or driver paid, disputes handled	Aging receivables; manual pay checks; disputes in email threads
Carrier sourcing	How new carriers/drivers are found, onboarded and kept	Spreadsheets of carriers; onboarding packets by email
Compliance	Files, expiries, pre-dispatch checks, incidents, audits	Expiry dates in someone's head; no pre-dispatch check; registers on paper

The eight things to record at every step

Record	Question to answer
Who	Which role does it (and who covers when they're out)?
System	Which screens or tools are open (TMS, email, spreadsheet, phone, portal)?
Time	How long one occurrence takes (median, and the bad case)
Frequency	How many times a day or week, and by how many people
Handoffs	Who it goes to next, and how (email, text, shout, shared sheet)
Re-keying	What gets copied from one system into another
Waits	Where the work sits waiting for a person, a reply or a document
Errors	What goes wrong, how often, and what it costs when it does

Hours per week = minutes per occurrence × occurrences per week × people ÷ 60. This number is the impact input for ranking (Section 6).

4. How to observe (the method)

Use the lightest method that answers the question. Most of the value comes from Week 1 and costs almost nothing.

When	Method	What it gives you
Week 0	Prep	Staff notice and opt-ins signed; pull 30 days of system timestamps (TMS load stages, email and phone counts) as a baseline. Set up client-owned storage.
Week 1	Shadowing, 2 to 3 hours per role	The real sequence and the pain in their words. Notes only, no recording of calls.
Week 1	Scribe captures (staff record their top 3 to 5 repeat tasks)	Exact click paths and systems per task. Doubles as SOPs the client keeps. Redact names, numbers and rates in screenshots.
Week 1	System timestamps	Volumes and cycle times without watching anyone (TMS stages, email/phone metadata, ELD events, spreadsheet edit history).
Week 2 (optional)	Opt-in browser extension on work browsers, volunteers only	Time per work site, tab switches, and copy-paste events between systems (never the content). Quantifies the re-keying shadowing only describes.
End	Readback	Findings to the participants first, then management.

Non-negotiable privacy rules

Written notice and opt-in for everyone, whatever the state. Opting out has no consequences.
No audio, no keystrokes, no typed text, no clipboard content, no passwords, no screenshots from the extension. Allow-listed work sites only; personal browsing and incognito are ignored; a visible pause button.
Process, not performance. Findings are reported by task and role, never ranked by person. Managers do not see raw per-person data.
The data is the client's. It goes to the client's own storage. Raw events are deleted 30 days after the pilot; only aggregated findings are kept.
Drivers and mobile phones are out of scope for device monitoring.
Counsel skims the staff notice before anything is deployed. All-party-consent states (Florida is one) mean never recording conversations without everyone's consent.
Skip enterprise task-mining suites for small teams. If the client already has Microsoft 365 with Power Automate Premium, its task mining can replace the extension, with screenshots off.

Questions to settle in Week 0

Google Workspace or Microsoft 365?
Who manages the PCs and browsers (needed to install the extension)?
Can we get ELD event logs and phone call-detail records (counts and durations only)?
Does the TMS keep a user activity or audit history we can export?

5. Readiness check

Before recommending any build, score the company on four areas. Anything marked red is the first recommendation, before any AI.

Area	Check	Green / amber / red
Data	Do the core records (loads, customers, carriers, invoices) live in one system of record?
	Are key fields filled in consistently (dims, weight, rates, status dates)?
	Can we get data out (API, export) and, where needed, back in?
	Who owns each data source, and who can fix it?
People and culture	Does the CEO visibly back this work?
	Is there a named owner for each pilot inside the team (not just me)?
	Have the skeptics been heard and invited in?
	Is there time for training, and a plan for people whose tasks change?
Risk and human approval	Are the "always human" decisions written down (prices, credit, payments, driver pay, safety, stop-work, hiring)?
	Is there an AI use policy and a register of AI tools and vendors?
	Are recording, consent and messaging rules known for the states involved?
	Are security basics in place (multi-factor sign-in, admin accounts known, no shared passwords)?
Fit with existing tools	Does a tool the company already pays for already do this? (Check before building.)
	Is this something to buy (TMS, ELD, DQ files, accounting) rather than build?
	Will the fix fit how people already work, or force a new screen on them?

Human-approval gates (always on, whatever gets built)

The system can draft, flag and remind. A named person approves: every price sent to a customer · customer credit · invoices released and factoring submissions · carrier or driver payments · carrier selection and onboarding · hiring decisions · safety, stop-work and return-to-duty · claims and incident closure · any message sent to customers or drivers in a new wording.

6. Ranking what you find

Score each candidate 1 to 5 on three things, where higher is always better, and multiply (maximum 125).

Score	5 means	1 means
I · Impact	Large observed hours, cash, or safety; hits a number leadership already tracks; several people feel it	Nice to have
E · Ease	Days of work, data access confirmed, fits existing tools	Months, or depends on data nobody can get
S · Safety	Read-only or draft-only, low legal and operational risk	Needs legal review, faces drivers or applicants, moves money, or carries vendor liability

Gates are separate from scores. A gate (contract, system access, legal review, insurer review, leadership sign-off) decides when, not how good.
Use observed hours from Section 3 for Impact. Replace estimates with observed numbers as soon as you have them, and re-rank.
Tag each item: G (generic, reusable method) or client (built from the client's own processes; stays theirs).
The top score isn't always first. The lowest-risk item that builds trust and produces evidence often goes first in the sequence.

Ranking sheet (fill in)

Candidate	Observed hrs/wk × people	I	E	S	Score	Gates	Human approval needed	G / client

7. Recommendation one-pager (template)

One page per recommendation. Written for the CEO: plain words, numbers, and a clear ask.

Section	Fill in
Title	[Short name, e.g. "Draft quotes from emailed requests"]
The problem, in their words	[Quote from an interview, with role, not name.] "..."
Evidence	[Observed hours/week × people; volumes; cycle times; errors and their cost. Say which numbers are observed and which are estimates.]
Who benefits and how	[Each affected role: what gets easier for them.]
Options	A. Do nothing: [cost of staying as is] B. Change the process / use a tool you already have: [ ] C. Buy: [product, price] D. Build: [what, effort]
Recommendation	[Option and why, in two sentences.]
Pilot	[Who, which loads or tasks, how many weeks, start date.]
Success metric	[One business number with today's baseline and the target, e.g. "quote turnaround from X hours to Y". Decision date to scale or stop.]
Cost	[One-time and monthly; client time needed.]
Risks and controls	[What could go wrong; human-approval points; data and privacy; what happens if it fails.]
Score	I [ ] × E [ ] × S [ ] = [ ] · Gates: [ ] · Tag: G / client
Approval	Approved by [name, title] · Date [ ] · Pilot owner [name]

Translating for the CEO (checklist)

☐ Lead with the business number, not the technology.
☐ Include "do nothing" and "buy" as real options.
☐ Say what you don't know yet.
☐ No promise the pilot can't measure.
☐ Name the people who helped shape it.

8. Discovery timeline and deliverables

When	Activities	Client receives
Week 0	Charter and contract signed; intake answered; CEO introduction; notice and opt-ins; baseline data pulled	Kickoff note: who we'll talk to and when; staff notice
Week 1	Interviews by role; shadowing; Scribe captures; system timestamps	Stakeholder map; first flow maps
Week 2	Optional opt-in extension; fill gaps; readiness check; first ranking	Readiness scorecard (green/amber/red)
Week 3	Readback to participants, then leadership; write one-pagers for the top 3	Findings report; ranked list; 3 recommendation one-pagers
Week 4	Leadership decision; pilot set up with an owner and success metric; AI policy and tool register drafted	90-day roadmap; AI use policy draft; AI tool and vendor register; SOPs from Scribe

A short engagement can stop at Week 3 (2 to 3 weeks total) with the findings report and ranked list. Timings are estimates and depend on access and people's availability.

Appendix A. Interview note sheet (one per person)

Role / team
Date, place, how long
What they're measured on
A normal day, in order
Systems open all day
Where they type the same thing twice
What they wait on
What goes wrong, and how often
Workarounds (sheets, notes, texts)
Worries about AI / change
What's in it for them
Magic wand answer (their words)
Willing to join a pilot?	Yes / Maybe / No
Follow-ups

Appendix B. Flow-step capture sheet

Flow / step	Who	System(s)	Min per time	Times / week	People	Hours / week	Hands off to (how)	Re-keyed (from → to)	Waits on	Errors / cost

Estimates and timings are illustrative, not commitments. Legal points are summaries, not legal advice. © Reshawn Holdings LLC d/b/a Voxelio AI. You're welcome to use and share this playbook with attribution.